add support for NaN fill values

[gdkrmr]

• Allows for using NaN, Inf and -Inf as fill values. Tested with my R wrappers only for NaN, I will test it for Inf tomorrow. Using NaN as fill value #148
• This PR also fixes a small bug (?): The raw compressor was missing from one of the maps between the compressor type and the string value.
• Not sure which NaN value to use, currently using std::numeric_limits<double>::quiet_NaN().
• No new unit tests.
• I cannot run the tests on my computer, because my OS is too old, so I simply transferred all the changes between files and did not run the tests.

[constantinpape]

Thanks for working on this and the changes look good to me.
For some reason it fails a few tests though.
I will investigate this later and let you know what I find.

[constantinpape]

There is one issue in the changes, I have fixed it in gdkrmr#1 and also
fixed another issue with the fillvalue and added a python unittest.
Once you merged that one, we can go ahead with this PR :).

[constantinpape]

Tests pass (never mind appveyor, the tests always fail on PRs for some reason) now and this looks good to me. Let me know if everything works out with your tests for +/- inf, then I will merge.

[gdkrmr]

Confirmed: Works from my side from R for NaNs, NAs, Inf, and -Inf 👍. I think this is ready to merge!

[constantinpape]

Thanks for the contribution :).